Table 5.
Top 5 important clinical features used by the EHR and combined models to identify etiologies of acute respiratory failure
| Diagnosis | EHR model | Combined model |
|---|---|---|
| Pneumonia | Oxygen saturation or PaO2 | Oxygen saturation or PaO2 |
| Procalcitonin | Procalcitonin | |
| Troponin-I | Troponin-I | |
| Absolute lymphocyte count | Plateau pressurea | |
| Plateau pressurea | BNP | |
| Heart failure | BUN or creatinine | BUN or creatinine |
| BNP | Troponin-I | |
| Troponin-I | BNP | |
| Tidal volumea | Tidal volumea | |
| Age | Age | |
| COPD | MCHC | MCHC |
| Oxygen saturation or PaO2 | Total bilirubin | |
| Lymphocytes % or neutrophils % | Bicarbonate | |
| Bicarbonate | Magnesium | |
| Age | Alkaline phosphate |
Note: Top features identified by permutation importance. Highly correlated features (>0.6) were grouped together during the permutation importance analysis and reported together (eg, BUN or Creatinine).
a
Plateau pressure and tidal volume measured during invasive mechanical ventilation.
BNP: brain natriuretic peptide; BUN: blood urea nitrogen; MCHC: mean corpuscular hemoglobin concentration; PaO2: partial pressure of oxygen.